Greetings,

I am working on an image processing plug-in (which I'll offer for upstream contribution BSD-licensed), and would really like to have a couple of primitives that I can't find.

One is finding "islands" in an image: take a channel and create new layers each with one contiguous region in the channel. So if the channel has ten islands, create ten new layers, each with one of the islands. Since this could be unwieldy (too many new layers), one new layer with one island would suffice; one could then process that island layer, subtract it from the original, and move on.

Another is integrals: zero, first and second moments, or alternatively, integrals/sums of channel intensity, x- and y-weighted intensity, and x^2, y^2 and xy-weighted intensity. This lets me fit an ellipse to an island, and then display major and minor axes.

Finally, opening and storing data files. I'd like to be able to create a CSV file with statistics of the islands: size, centroid, aspect ratio (major/minor axis ratio), major axis orientation. One could then further process these statistics in a spreadsheet or other program.

I can do the first two in scheme, though it would be slow, but tolerable for my needs. I don't see a script-fu primitive which would let me do the last.

If there are other packages which can do this, that would be helpful. But having it in the same program/UI as GIMP, with all of its other image manipulation tools, and with its presence across platforms, would be even better. I'd love to have your ideas.

Cheers, -Adam

On Sat, 2008-11-01 at 01:02 +1030, David Gowers wrote:

Hi Adam,

On Fri, Oct 31, 2008 at 9:51 PM, Adam C Powell IV wrote:

Greetings,

I am working on an image processing plug-in (which I'll offer for upstream contribution BSD-licensed), and would really like to have a couple of primitives that I can't find.

One is finding "islands" in an image: take a channel and create new layers each with one contiguous region in the channel. So if the channel has ten islands, create ten new layers, each with one of the islands. Since this could be unwieldy (too many new layers), one new layer with one island would suffice; one could then process that island layer, subtract it from the original, and move on.

Another is integrals: zero, first and second moments, or alternatively, integrals/sums of channel intensity, x- and y-weighted intensity, and x^2, y^2 and xy-weighted intensity. This lets me fit an ellipse to an island, and then display major and minor axes.

Finally, opening and storing data files. I'd like to be able to create a CSV file with statistics of the islands: size, centroid, aspect ratio (major/minor axis ratio), major axis orientation. One could then further process these statistics in a spreadsheet or other program.

I can do the first two in scheme, though it would be slow, but tolerable for my needs. I don't see a script-fu primitive which would let me do the last.

If there are other packages which can do this, that would be helpful. But having it in the same program/UI as GIMP, with all of its other image manipulation tools, and with its presence across platforms, would be even better. I'd love to have your ideas.

It sounds like you might find working with the combination of Gimp-Python, NumPy and SciPy more effective. integration,definitely. Islands, possibly (a browse through http://scipy.org/scipy/scipy/browser/trunk might help you to find out. And Python has a CSV module as standard.

Okay, thanks for the suggestion. The downsides seem to be: I and others need to install these as well as GIMP to use it, I need to learn Python (yes, I'm one of the tiny handful remaining who know scheme but not python :-), and converting from a GIMP image to something which NumPy/SciPy can quickly sum might not be straightforward.

I had the impression that writing external files was not supported by Script-Fu; I'm sure Kevin Cozens could tell you.

Okay, I look forward to hearing from him. Hmm, I suppose if the output is small enough, this could return it as a list, and one could run it in the script-fu console then copy-and-paste the return value...

Hope that helps,

Yes, thanks much.

-Adam

On Fri, 2008-10-31 at 19:27 -0400, Kevin Cozens wrote:

David Gowers wrote:

One is finding "islands" in an image

[snip]

Another is integrals: zero, first and second moments, or alternatively, integrals/sums of channel intensity, x- and y-weighted intensity, and x^2, y^2 and xy-weighted intensity.

As you stated, the above two features could be implemented in Scheme but they would probably be too slow in actual use.

Indeed, just tried, they're quite slow. When I can't bear it any longer, I'll see about a C plugin.

Finally, opening and storing data files.

See section 6.6 of the R5RS. Working with external files is handled via the use of pipes.

Thanks, got it working. And Saul, text is all I need, can't speak for others.

I don't suppose there's a script-fu interface to the file browser for setting the file name...

One more question: I can't seem to access procedures in my scripts from the console. That is, if I define some procedures in files in .gimp-2.4/scripts/foo, the console doesn't know they exist. Is there a way to access those procedures from the console?

Cheers, -Adam